What’s up everyone. I have a question in regards to some work I’m doing. I need to get the total accuracy.

Here is the problem:

I have -

Expected Number
Actual Number

The difference between the expected and actual number are both positive and negative variances

Currently I turned both positive and negative numbers to absolute values.

The formula I am using = 1 - (variance(Absolute)/expected number) * 100%

Is this the correct way to get the total accuracy?

Depends on what you mean by accuracy. Variance is usually a measure of accuracy. You could also use kurtosis to measure how big the tails are, which are further from the average.

(actual-expected)/expected

example below...
given: actual=75, expected=68

(75-68)/68=.10 or 10% meaning you were 10% above expected

depending on how you're treating the "expected" value (i.e. if "expected" means something negative (like failures/hour or smth) then you may want to reflect the values negatively instead (meaning you failed 10% more than expected) by correcting the formula to

-(actual-expected)/expected

This post was edited by Penguins0690 on Aug 19 2023 08:24pm

Yeah so the variance is the accuracy based off actual/expected. The issue I’m having is the variance can be both negative and positive. For example.

We have a table:

Expected l Actual l Variance l Absolute variance
100 l 88 l -12 l 12
75 l 67 l - 8 l 8


The formula I am using is
1 - (Absolute variance/expected) * 100%

Now what I do in power query is group them so it would be a table like this

Expected l Actual l Variance l Absolute variance
175 l 155 l -20 l 20




Another issue I am having is if the absolute is more then the expected I am getting a negative percentage back…

For example:

Expected actual variance absolute variance
12 24 12 12

This post was edited by 300 on Sep 11 2023 12:13pm

Variance is something different (the mean squared deviation).

You would divide by the actual number (assuming you want to ascertain how accurate the estimate of the actual number by the expected number was). And the absolute value would have to be taken after subtracting the fraction from 1, not before.

expected 30, actual 90 is a 66.6% deviation from the actual number. abs(30/90-1)=two thirds
expected 100, actual 20 is a 400% deviation from the actual number. abs(100/20-1)=4

abs(expected/actual -1) = abs((expected-actual)/actual)

Check this out: https://en.wikipedia.org/wiki/Mean_absolute_percentage_error

From the standard deviation wiki basic example it looks like, to find the variance you:
First calculate the error of each (actual - expected).
Then find the total average error.
Now, subtract the average from each error and square the result.
Find the average of the results.

But, it sounds like you really want a goodness of fit calculation, to check whether your estimates are accurate.

This post was edited by effinA on Sep 11 2023 02:53pm